MCPcopy Create free account

hub / github.com/deepspeedai/DeepSpeed / functions

Functions11,258 in github.com/deepspeedai/DeepSpeed

⨍ Functions 11,258◇ Types & classes 1,887↳ Endpoints 34
↓ 3 callersFunction_expert_weight_parity_worker
(rank, world_size, tp_size, ep_size, zero_stage=0)
tests/unit/v1/moe/test_autoep_autotp_grad_parity.py:605
↓ 3 callersFunction_fake_modal
( candidate_sha: str, fail_label: str | None = None, cleanup_failure: bool = False, wait_failu
ci/test_torch_latest.py:132
↓ 3 callersFunction_folded_zero2_config
(*, mixed_precision=True)
tests/unit/v1/moe/test_autoep_autotp_grad_parity.py:72
↓ 3 callersMethod_fp32_state_allgather
(self, param, fp32_state_partition)
deepspeed/runtime/zero/stage3.py:2770
↓ 3 callersMethod_fuse_lora_layer
(self, layer_id)
deepspeed/runtime/hybrid_engine.py:132
↓ 3 callersFunction_gather_optimizer_state_for_param
(engine, param, key)
tests/unit/v1/moe/test_autoep_checkpoint.py:199
↓ 3 callersFunction_gather_source_zero_params
Gather source ZeRO params while AutoEP reads full tensor values.
deepspeed/moe/ep_repack.py:25
↓ 3 callersFunction_gather_tokens
Gather tensors and concatenate them along a dimension
deepspeed/moe/mappings.py:30
↓ 3 callersFunction_gather_zero_param
(param)
tests/unit/v1/moe/test_autoep_checkpoint.py:150
↓ 3 callersFunction_get_autoep_metadata
(model_state)
deepspeed/checkpoint/ds_to_universal.py:497
↓ 3 callersMethod_get_buffer
(self, index)
deepspeed/runtime/swap_tensor/async_swapper.py:157
↓ 3 callersMethod_get_ckpt_name
(self, checkpoints_path, tag, mp_placeholder=None, pp_placeholder=None)
deepspeed/runtime/engine.py:4146
↓ 3 callersMethod_get_current_buffer
(self)
deepspeed/runtime/swap_tensor/utils.py:173
↓ 3 callersFunction_get_data_parallel_group
Get the data parallel group the caller rank belongs to.
deepspeed/utils/groups.py:701
↓ 3 callersFunction_get_expert_data_parallel_group
Get the expert data parallel group the caller rank belongs to.
deepspeed/utils/groups.py:638
↓ 3 callersFunction_get_expert_parallel_group
Get the expert parallel group the caller rank belongs to.
deepspeed/utils/groups.py:619
↓ 3 callersFunction_get_file_path
(tmpdir, file_prefix, index=0)
tests/unit/v1/nvme/test_aio.py:38
↓ 3 callersFunction_get_fw_real_inputs
Resolve graph-local real inputs from the one-shot queue or persistent storage.
deepspeed/compile/backend.py:227
↓ 3 callersFunction_get_local_rank
()
tests/unit/v1/nvme/test_gds.py:26
↓ 3 callersMethod_get_optimizer_param
(self, param_name)
deepspeed/runtime/engine.py:3562
↓ 3 callersMethod_get_optimizer_state
(self, sd, state_key)
deepspeed/checkpoint/zero_checkpoint.py:120
↓ 3 callersFunction_get_param_uc_restore_meta
Return the restore-facing view of AutoTP UC metadata for a parameter. AutoTP parameter metadata intentionally serves two separate consumers:
deepspeed/checkpoint/universal_checkpoint.py:23
↓ 3 callersFunction_get_test_write_file_and_device_buffer
(tmpdir, ref_buffer, gds_handle, index=0)
tests/unit/v1/nvme/test_gds.py:52
↓ 3 callersFunction_get_test_write_file_and_pinned_tensor
(tmpdir, ref_buffer, aio_handle=None, index=0)
tests/unit/v1/nvme/test_aio.py:62
↓ 3 callersFunction_get_test_write_file_and_unpinned_tensor
(tmpdir, ref_buffer, index=0)
tests/unit/v1/nvme/test_aio.py:56
↓ 3 callersMethod_get_universal_checkpoint_info
(self)
deepspeed/runtime/bf16_optimizer.py:243
↓ 3 callersMethod_get_used_buffers
(self)
deepspeed/runtime/swap_tensor/utils.py:176
↓ 3 callersFunction_get_zero_param_intra_parallel_group
Get the ZeRO parameter partitioning intra parallel group the caller rank belongs to.
deepspeed/utils/groups.py:889
↓ 3 callersMethod_git
(self, *args: str)
ci/test_tests_fetcher.py:87
↓ 3 callersFunction_git_command
(*args: str)
ci/torch_latest.py:214
↓ 3 callersFunction_group_per_tp_shapes
(slice_shapes_by_tp, pp_degree, tp_degree)
deepspeed/checkpoint/ds_to_universal.py:726
↓ 3 callersMethod_has_inf_or_nan
(x, j=None)
deepspeed/runtime/zero/stage_1_and_2.py:2423
↓ 3 callersFunction_init_engine
(config_dict, hidden_dim, seed=42)
tests/unit/v1/zero/test_zero2_offload_multi_backward.py:43
↓ 3 callersFunction_init_group_wise_weight_quantization
[Experimental] Apply group-wise weight quantization to model. Replace layers module according to config_list Args: model (nn.Module): A n
deepspeed/inference/quantization/quantization.py:20
↓ 3 callersMethod_is_grid_valid
(self)
deepspeed/runtime/pipe/topology.py:390
↓ 3 callersMethod_is_partition_offloaded
Whether the parameter partition needs the per-parameter offload path. The selective update assumes the partition is resident on the compute d
deepspeed/ops/adam/zenflow_torch_adam.py:403
↓ 3 callersMethod_link_all_hp_params
(self)
deepspeed/runtime/zero/stage_1_and_2.py:717
↓ 3 callersFunction_load_universal_dense_state
(universal_dir, param_name, key)
tests/unit/v1/moe/test_autoep_checkpoint.py:60
↓ 3 callersFunction_load_universal_expert_state
(universal_dir, param_name, key)
tests/unit/v1/moe/test_autoep_checkpoint.py:66
↓ 3 callersFunction_make_autoep_zero2_config
(ep_size)
tests/unit/v1/moe/test_autoep_grad_parity.py:50
↓ 3 callersFunction_make_freqs
(seq_len, rot_dim, theta=10000.0, device="cpu")
tests/unit/sequence/test_apply_rotary_pos_emb.py:13
↓ 3 callersMethod_make_key
(self, i, j)
deepspeed/checkpoint/reshape_meg_2d.py:52
↓ 3 callersFunction_make_spec
(**kwargs)
tests/unit/v1/moe/test_autoep_unit.py:69
↓ 3 callersFunction_mark_fake_zero_param
(param, full_data, partition_data=None, ds_id=0, name="param")
tests/unit/v1/moe/test_autoep_unit.py:104
↓ 3 callersMethod_mlp_gemm
(input, residual, input_bias, weight_interm, weight_out, bias, gamma, beta, eps, pre_layer_norm,
op_builder/npu/inference.py:214
↓ 3 callersMethod_model_parallel_all_reduce
Perform all reduce within model parallel group, if any.
deepspeed/runtime/zero/stage3.py:2186
↓ 3 callersMethod_model_parallel_all_reduce
Perform all reduce within model parallel group, if any.
deepspeed/runtime/zero/stage_1_and_2.py:1984
↓ 3 callersFunction_new_handle
()
tests/unit/v1/nvme/test_pinned_manager.py:20
↓ 3 callersFunction_offloaded_stage3_param
(selected_indices)
tests/unit/ops/adam/test_zf_torch_adam.py:227
↓ 3 callersFunction_one_hot_to_float
(x, num_classes)
deepspeed/moe/sharded_moe.py:180
↓ 3 callersMethod_optimizer_has_ckpt_event_epilogue
(self)
deepspeed/runtime/engine.py:1500
↓ 3 callersMethod_optimizer_has_ckpt_event_prologue
(self)
deepspeed/runtime/engine.py:1497
↓ 3 callersMethod_overflow_check_and_loss_scale_update
(self)
deepspeed/runtime/zero/stage3.py:2493
↓ 3 callersFunction_patch_kwargs
(kwargs, monkeypatch)
tests/unit/compile/test_inductor_aot_kwargs.py:51
↓ 3 callersMethod_placeholder
(self, param)
tests/unit/ops/adam/test_zf_torch_adam.py:204
↓ 3 callersFunction_pre_ln_test_helper
(n_tokens: int, n_channels: int, dtype: torch.dtype, res_add: bool = False)
tests/unit/inference/v2/modules/test_cuda_pre_ln_module.py:36
↓ 3 callersFunction_pre_rms_test_helper
(n_tokens: int, n_channels: int, dtype: torch.dtype, res_add: bool = False)
tests/unit/inference/v2/modules/test_pre_rms_module.py:38
↓ 3 callersMethod_pre_step
(self)
deepspeed/runtime/zero/stage3.py:2311
↓ 3 callersMethod_precompute_checkpointable_values
(self)
deepspeed/runtime/pipe/module.py:222
↓ 3 callersMethod_qkv_gemm
(inputs, weight, q_scale, bias, gamma, beta, eps, add_bias, q_int8, transpose)
op_builder/npu/inference.py:58
↓ 3 callersFunction_quantize_module
(model, policies)
deepspeed/module_inject/module_quantize.py:63
↓ 3 callersMethod_raise_if_autoep_zero3_consolidated_export
(self, operation)
deepspeed/runtime/engine.py:5403
↓ 3 callersMethod_reassign_or_swap_out_partitioned_parameters_async
Asynchronously update partitioned parameters with optimized values.
deepspeed/runtime/superoffload/superoffload_stage3.py:129
↓ 3 callersMethod_recursive_getattr
Allow getting an attribute like "linear.weight
deepspeed/runtime/pipe/module.py:450
↓ 3 callersMethod_release_tensor
(self, tensor_id)
deepspeed/runtime/zero/contiguous_memory_allocator.py:147
↓ 3 callersMethod_replace_with_config
Replace layer using the new configurable AutoTP API. This method uses TPLayerSpec to determine how to partition the layer.
deepspeed/module_inject/auto_tp.py:414
↓ 3 callersFunction_require_aio_cuda
()
tests/unit/v1/nvme/test_aio_gil.py:25
↓ 3 callersMethod_residual_add_bias
(hidden_state, residual, attention_output, attention_bias, final_bias, mp_size, mlp
op_builder/npu/inference.py:255
↓ 3 callersFunction_router_params
(engine)
tests/unit/v1/moe/test_autoep_checkpoint.py:135
↓ 3 callersFunction_run_folded_checkpoint_same_topology_resume
(checkpoint_dir, *, ep_size=2, mixed_precision=True)
tests/unit/v1/moe/test_autoep_autotp_checkpoint.py:256
↓ 3 callersFunction_run_multi_backward
(config_dict, hidden_dim, num_chunks, num_steps=1, seed=42)
tests/unit/v1/zero/test_zero2_offload_multi_backward.py:64
↓ 3 callersMethod_run_test
(self, tmpdir, dtype, ds_config, load_optim, use_torch_adam, world_size)
tests/unit/checkpoint/test_universal_checkpoint.py:216
↓ 3 callersMethod_set_client_model
(self, model)
deepspeed/runtime/engine.py:1693
↓ 3 callersFunction_set_module_by_name
Replace the submodule at *dotted_name* with *new_module* in-place.
deepspeed/sequence/auto_sp.py:134
↓ 3 callersMethod_setup_baseline_linears
(self, model)
tests/unit/model_parallelism/test_tp_plan_e2e.py:53
↓ 3 callersMethod_softmax_context
(query_key_value, attn_mask, rotary_dim, rotate_half, rotate_every_two, heads, num_kv,
op_builder/npu/inference.py:114
↓ 3 callersFunction_source_param_shape
(param: torch.Tensor | nn.Parameter)
deepspeed/module_inject/auto_ep.py:85
↓ 3 callersFunction_split_plan_counts
(scenario)
tests/unit/v1/moe/test_autoep_unit.py:910
↓ 3 callersMethod_stop_timers
(self, timer_names)
deepspeed/runtime/engine.py:3489
↓ 3 callersMethod_sycl_env_paths
Find the SYCL include and lib directories from the Python environment. When using PyTorch XPU wheels, libsycl.so and SYCL headers are
op_builder/xpu/builder.py:51
↓ 3 callersMethod_sync_cpu_optimizer_step
(self, param_group_id: int, sub_group_id: in
deepspeed/runtime/superoffload/superoffload_stage3.py:314
↓ 3 callersMethod_test_set_z3_leaf_modules
(self, cls, requires_grad)
tests/unit/runtime/zero/test_zero_leaf_module.py:327
↓ 3 callersMethod_unflatten_partitioned_parameters
(self, sub_group_id)
deepspeed/runtime/zero/stage3.py:2455
↓ 3 callersMethod_update_hp_grad
(self, lp, group_idx, param_idx, clear_lp_grads)
deepspeed/runtime/bf16_optimizer.py:356
↓ 3 callersMethod_update_param_state_info
(self, swap_info, parameter)
deepspeed/runtime/swap_tensor/optimizer_utils.py:490
↓ 3 callersMethod_valid_micro_batch
(self, micro_batch_id)
deepspeed/runtime/pipe/schedule.py:80
↓ 3 callersMethod_validate_autoep_zero3_partitioned_metadata
(autoep_layers, model=None, require_partitioned=True)
deepspeed/runtime/engine.py:4309
↓ 3 callersMethod_vector_matmul
(input, weight, async_op, q_scale, q_int8, transposed_mode)
op_builder/npu/inference.py:196
↓ 3 callersFunction_verify_continuous_decrease
(values)
tests/unit/runtime/test_lr_schedulers.py:23
↓ 3 callersMethod_wildcard_to_regex
(pattern: str)
deepspeed/module_inject/tp_plan_converter.py:62
↓ 3 callersMethod_writeback_swap_state
(self, sub_group_id, write_opt_state, write_gradients)
deepspeed/runtime/zero/stage3.py:2450
↓ 3 callersMethod_zero3_partition_group_metadata
(self)
deepspeed/runtime/zero/stage3.py:3093
↓ 3 callersFunction_zero_optimizer_stub
(*, partition_gradients, overlap_comm)
tests/unit/v1/moe/test_autoep_autotp_zero1_overlap_folding.py:15
↓ 3 callersMethodabsolute_name
Returns absolute build path for cases where the op is pre-installed, e.g., deepspeed.ops.adam.cpu_adam will be installed as something
op_builder/builder.py:132
↓ 3 callersFunctionadam_update
(grad, buf1, buf2, step, betas, eps)
deepspeed/runtime/zero/muon/original_muon.py:267
↓ 3 callersFunctionadd_end_backward
(graph: Graph, graph_id: int, release_reduce_buckets: bool = True)
deepspeed/compile/fx.py:28
↓ 3 callersMethodadd_pointer_offset
Adds a pointer offset in units of Element
csrc/deepspeed4science/evoformer_attn/iterators/predicated_tile_iterator_atomic.h:330
↓ 3 callersMethodall_gather_coalesced
(params: Iterable[Parameter], safe_mode: bool = False,
deepspeed/runtime/zero/partition_parameters.py:1495
↓ 3 callersFunctionall_to_all_quant_reduce
(tensors: List[Tensor], groups: {})
deepspeed/runtime/comm/coalesced_collectives.py:31
↓ 3 callersFunctionallclose
(x, y)
tests/unit/ops/spatial/test_nhwc_bias_add.py:17
← previousnext →901–1,000 of 11,258, ranked by callers